Rune 26B-A4B
Invergent's decision model for text and images. A fine-tune of Gemma-4-26B-A4B (about 4B parameters active per token) that answers Choice, Noul and Score questions, with an optional thinking mode for harder questions.
The Hugging Face repository is gated (access must be requested) and, despite its name, holds BF16 safetensors shards; a hosted API is also offered. Invergent reports a Decision Index 0.2.1 score of 57.44 against Jev 1.13's 57.91, ahead of Jev in three of the five areas, and 90 to 180 ms per decision on an RTX PRO 6000 (0.18 s with an image, 4.9 s with thinking). On the public Decision Index board it was first among open models on 28 September. Context 262,000 tokens. Weights Apache-2.0, but check the gate's terms before mirroring.
What it decides
- choice — picks one option from a set
- score — places the input on an ordered scale
- noul — answers a yes/no question with one calibrated probability
- classify — assigns a category from a fixed taxonomy
- route — sends the input to one of several destinations
At a glance
| Parameters | 26B |
| Base model | google/gemma-4-26b-a4b-it |
| Maker | Surogate (Invergent) |
| Released | 2026-09-21 |
| License | apache-2.0 |
| Reported latency | 90–180 ms per decision on an RTX PRO 6000; 0.18 s with an image; 4.9 s with thinking |
Hosted API
Served by Invergent. Get access.
Get the weights
pip install systemonemodels
systemone pull surogate/rune
The files are served from the maker's Hugging Face repository, surogate/rune-26b-a4b-GGUF, and verified against the checksums recorded here.